C3
CalibrationBase Rate Neglect
Prior probabilities should be weighted appropriately
4
Models Tested
1
Confirmatory
-0.362
Mean Effect
0.501
Max Effect
Theoretical Context
Theoretical Anchor
Kahneman & Tversky (1973)
Normative Violation
Prior probabilities should be weighted appropriately
Cross-Model Comparison
Effect sizes for Base Rate Neglect across all tested models
Anthropic
Claude Sonnet 4.6
The Cautious Contrarian
h = -0.501Confirmatory
Google
Gemini 2.0 Flash
The Consistent Optimist
h = -0.384
OpenAI
GPT-5.3 Instant
The Directive Optimist
h = -0.371
OpenAI
GPT-5.2
The Steady Traditionalist
h = -0.194
Statistical Details
Full results with confidence intervals and sample sizes
| Model | n (A) | n (B) | Cohen's h | 95% CI | Status |
|---|---|---|---|---|---|
| Claude Sonnet 4.6 | 50 | 50 | -0.5010 | — | Confirmatory |
| Gemini 2.0 Flash | 50 | 50 | -0.3835 | — | Exploratory |
| GPT-5.3 Instant | 50 | 50 | -0.3714 | — | Exploratory |
| GPT-5.2 | 50 | 50 | -0.1935 | — | Exploratory |